BCCard/bc-finance-llm-golden-set
收藏资源简介:
BC Card Finance LLM Golden Set 是一个用于LLM部署前回归测试和质量验证的黄金集数据集。它是BC카드-연세대 DSL 산학협력(S2026 LLMOps项目)的产出物。数据集包含三个主要文件:data/regression_golden_set.csv(用于回归测试的核心黄金集,共300行)、data/financial_edge_cases_200.csv(用于检测金融LLM幻觉的边缘案例,共200行)和data/hard_negatives_qa_50.csv(硬负样本,共50行)。该数据集主要用于模型更新或提示变更后的部署前回归测试,如果分数低于阈值,则会阻止部署。许可证为Apache 2.0。
BC Card Finance LLM Golden Set is a golden dataset intended for pre-deployment regression testing and quality validation of Large Language Models (LLMs). It is the deliverable of the BC Card-Yonsei University DSL Industry-Academia Cooperation (S2026 LLMOps Project). The dataset includes three primary files: data/regression_golden_set.csv, the core golden set for regression testing with 300 rows in total; data/financial_edge_cases_200.csv, a set of edge cases for detecting hallucinations in financial LLMs with 200 rows in total; and data/hard_negatives_qa_50.csv, a collection of hard negative QA samples with 50 rows in total. This dataset is mainly used for pre-deployment regression testing after model updates or prompt modifications. If the evaluation score falls below the predefined threshold, deployment will be blocked. The dataset is licensed under Apache 2.0.




